摘要
目的提出一种改进的模糊C均值聚类算法,并将其应用于脑部MR图像分割。方法首先,采用最大距离测度选取FCM初始聚类中心;然后,采用硬分类方法更新聚类中心和空间邻域信息构建隶属度函数,最后完成图像各区域分类。结果采用人工合成图像和临床实例脑部MR图像进行仿真实验,结果显示基于空间信息的SFCM/SFFCM算法所得图像噪声水平低于传统的FCM/FFCM算法。定量分析显示基于SFCM1,1/SFFCM1,1的分类评价指标模糊位置系数Vpc(0.944)和位置信息熵Vpe(0.043)均最优,SFFCM1,1程序耗时较标准FCM降低了37.2%~82.9%,迭代次数减少5~20次。结论本研究提出的SFFCM分割算法收敛速度更快,精确度更高,是一种可行的脑部MR图像分割算法。
Objective This paper proposed an improved fuzzy C-mean clustering(FCM)algorithm which was used for the segmentation of brain MR image.Methods Firstly,the maximum distance measure was used to select the initial clustering center of FCM.Then,the membership function was constructed by updating the clustering center and spatial neighborhood information with hard classification method.Finally,the image regions were classified.Results Artificial synthetic images and clinical brain MR images were used for experiments.The results showed that the image noise level of SFCM/SFFCM algorithm based on spatial information was lower than that of traditional FCM/FFCM algorithm.Quantitative analysis showed that the fuzzy partition coefficient Vpc(0.944)and partition entropy Vpe(0.043)of classification evaluation index based on SFCM1,1/SFFCM1,1 were optimal.Compared with standard FCM,program of SFFCM1,1 consumed 37.2%~82.9%less time and reduced the number of iterations by 5~20 times.Conclusion The SFFCM segmentation algorithm proposed in this study has faster convergence speed and higher accuracy,which is a feasible brain MR image segmentation algorithm.
作者
任彤
REN Tong(Department of Radiology,Nanjing Chest Hospital,Nanjing Jiangsu 210029,China)
出处
《中国医疗设备》
2019年第9期93-95,109,共4页
China Medical Devices
关键词
模糊C均值
空间邻域信息
脑部磁共振
图像分割
fuzzy C-mean
spatial neighborhood information
brain MR
image segmentation
